Epidemiology Class Notes 2012-09-27

In the News

s = rpois(10000,lambda=250)
change = diff(s)/s[1:9999]
require(lattice)
## Loading required package: lattice
densityplot(change)

table( change > .23)
## 
## FALSE  TRUE 
##  9898   101

Estimation Problem 1: Number of Physicians

22000 doctors A multiplication problem in converting patients per week to an annual basis. 708000 doctors Note 4 doctors per person. 600000 doctors and 4000 slots per medical school Using established rates of doctors per 1000 to assess “need.” Medical school estimate doesn’t take into account the working life of doctors. 200000 slots a year in medical school Note the big career drop out rate. *60000 slots per year 10% retirement rate.

AAMC document See page 11 for fall in number of physicians after Flexner report (175 to 125 per 100000). Figure 3 on page 12 for number of medical school graduates. Total physicians, see page 17: about 800000

Linking Exposure and Disease

An example of risk ratios and population attributable fraction

Research article on sitting and mortality * Work through abstract * p. 1 Definition of activity metabolic equivalents * p. 1 Why “prospective” studies? How would you do an experiment about “television viewing”, “sitting in a car”, etc. * p. 2 Potential confounding variables: sex, age, education level, … * p. 2 “All … analysis were adjusted for sex, age, ….” * Table 1. Who sits a lot? People in poor health. People with little physical activity.
* Table 2. Note “dose/response”. Note consistency across categories. Oddball: 150-299 mn/wk of physical activity as a low Hazard ration for sitting > 11 hr per day. But look at the confidence interval. * Figure. Activity is good, sitting is bad, for both people in good health and people in bad health.

Rate ratios versus absolute difference in rate

Table 5.1

Compare “never smoked” to “current smoker”. Rate ratio is 2.8 = 49.6/17.7. But the absolute difference in rate is 49.6-17.7 incidence rate per 100,000 person years.

Relative Risk

Just like rate ratios but based on cumulative incidence. Example

Table 1.2

Another example: examining the affect of an intervention (Calling people to remind them of an immunization):

Table 5.2

Relative Risk Reduction

Box 5.2 showing affect of pressure bandages in patients undergoing coronary angiography.

Attributable Risk

A graphic on attributable risk: Fig 5.1 Use this to derive the attributable fraction as \(latex \frac{RR-1}{RR}\)

In the British Doctors Study: Table 5.4

Population Attributable Risk

Combine the rates of the different groups and compare to the background rate. Fig. 5.2 Smoking and Tuberculosis in India. See the World Health Organization Exercise * WHO spreadsheet here

Case Control

Why we do them? When the incidence/prevalence is very small, so a cohort is not efficient.

What’s an “odds”? From “Fugue for Tinhorns”
I’m pickin’ Valentine, ’cause on the morning line
A guy has got him figured at five to nine

Example: Ovarian cancer and oral contraception use: Table 5.6

Example: Bicycle helmets and head injury Table 5.9

Why the Odds Ratio Approximates the Relative Risk (when the prevalence/incidence is very small): Box 5.5

Odds Ratios

  • Don’t reject the odds ratio. It’s needed for logistic regression which is needed to handle confounding. See paragraph above “Summing Up” in Making Sense of Odds Ratios > “… odds ratios are the output of logistic regression [we’ll use stratification], a technique often used to control for confounding bias in research analysis. Adjustment of odds ratios in logistic regression analysis is widely available through software packages. [We do it in Math 155.] Properly addressing confounding in research usually trumps concerns regarding odds ratios.”
  • Make decisions based on “Number Needed to Treat” and “Number Needed to Harm”. See Down with Odds Ratios. Story about Humira doubling the risk of lymphoma. But the risk of lymphoma is small compared to the risk of aggrevation of the disease being treated by Humira.

Conversion between OR and event rate (ER) and Number Needed to Treat

## 
## Attaching package: 'dplyr'
## The following objects are masked from 'package:stats':
## 
##     filter, lag
## The following objects are masked from 'package:base':
## 
##     intersect, setdiff, setequal, union
## 
## Attaching package: 'ggstance'
## The following objects are masked from 'package:ggplot2':
## 
##     geom_errorbarh, GeomErrorbarh
## 
## New to ggformula?  Try the tutorials: 
##  learnr::run_tutorial("introduction", package = "ggformula")
##  learnr::run_tutorial("refining", package = "ggformula")
## Registered S3 method overwritten by 'mosaic':
##   method                           from   
##   fortify.SpatialPolygonsDataFrame ggplot2
## 
## The 'mosaic' package masks several functions from core packages in order to add 
## additional features.  The original behavior of these functions should not be affected by this.
## 
## Note: If you use the Matrix package, be sure to load it BEFORE loading mosaic.
## 
## Attaching package: 'mosaic'
## The following object is masked from 'package:Matrix':
## 
##     mean
## The following object is masked from 'package:ggplot2':
## 
##     stat
## The following objects are masked from 'package:dplyr':
## 
##     count, do, tally
## The following objects are masked from 'package:stats':
## 
##     binom.test, cor, cor.test, cov, fivenum, IQR, median, prop.test,
##     quantile, sd, t.test, var
## The following objects are masked from 'package:base':
## 
##     max, mean, min, prod, range, sample, sum
plotFun( log2((1 - ER*(1-OR))/(ER*(1-OR)*(1-ER)))~ OR&ER, 
         OR.lim=c(0.5,0.9), ER.lim=c(0.5, 0.9), main="log2(NNT)")